This repository is for GOSS Segmentor (GST) introduced in the following paper:
"GOSS: Towards Generalized Open-set Semantic Segmentation"
If you find our paper/code is useful, please cite:
@article{hong2024goss,
title={Goss: Towards generalized open-set semantic segmentation},
author={Hong, Jie and Li, Weihao and Han, Junlin and Zheng, Jiyang and Fang, Pengfei and Harandi, Mehrtash and Petersson, Lars},
journal={The Visual Computer},
volume={40},
number={4},
pages={2391--2404},
year={2024},
publisher={Springer}
}
- Install packages in
README_pkgs.md
- If you had the error of "undefined symbol" as importing
pycocotools
, use the following steps:
pip uninstall pycocotools
git clone https://github.com/pdollar/coco.git
cd ./coco/PythonAPI
CC=gcc python3 setup.py build_ext install
You could download datasets from here. We expect the directory structure to be the following:
datasets/coco
train2017/ # train images
val2017/ # val images
datasets/coco_stuff_voc
annotations/
segments_voc_20_60_train2017.json
segments_voc_20_60_val2017.json
voc_20_60/
voc_20_60_train2017/
voc_20_60_val2017/
datasets/coco_stuff_manual
annotations/
segments_manual_20_60_train2017.json
segments_manual_20_60_val2017.json
manual_20_60/
manual_20_60_train2017/
manual_20_60_val2017/
datasets/coco_stuff_random
annotations/
segments_random_20_60_train2017.json
segments_random_20_60_val2017.json
random_20_60/
random_20_60_train2017/
random_20_60_val2017/